Hybrid Domain Joint Low-Rank and Sparse Constrained Framework for Simultaneous RFI Separation and SAR Imaging

The inherent contradiction between limited spectrum resources and ever-growing spectrum demand makes it difficult for synthetic aperture radar (SAR) to avoid in-band radio frequency interference (RFI) during observation, leading to significant degradation in imaging quality. In this context, RFI suppression technology has become crucial for obtaining high-fidelity SAR images. Existing RFI suppression methods are primarily divided into pre-processing and post-processing categories, both of which have inherent limitations. Pre-processing methods often struggle to achieve sufficient suppression due to the high overlap between target echoes and RFI in the time, frequency, time-frequency and other domains. Post-processing methods, which operate directly on the imaging results, tend to cause loss of image details and weak targets. Motivated by these problems, this paper proposes a hybrid domain joint lowrank and sparse constrained framework for simultaneous RFI separation and SAR imaging. The innovation lies in leveraging the low-rank characteristic of RFI in the echo domain as well as the low-rank and sparse representation of scene targets, while coupling the echo and image domains through the SAR imaging operator to establish a unified inversion model with relational constraints. To solve this model, a deep learning network named SSI-net is constructed by integrating the alternating direction method of multipliers with deep unfolding techniques. Bedisdes, for scenarios with limited or no training samples, a gradient descent-based optimization algorithm is also proposed. Simulations and experiments with measured data demonstrate the superior performance of the proposed framework.